Chapter 14

Using AI Intentionally

The Complete Framework, Tool by Tool

“A tool is not good or bad in itself. What matters is what you do with it and what it does to you.”

— After Ivan Illich

Everything in this book compresses into one question, asked before any significant AI interaction: am I about to extend my thinking, or replace it? This chapter turns that question into an operating framework — four steps, then their application to the four tool categories of a normal professional week. If you keep nothing else from these pages, keep this chapter; the rest of the book is its argument.

The Four Steps

1 — Engage first

Before the tool, spend genuine time with the problem — twenty minutes, a rough draft, a position brief. Not to finish before getting help; to prime the engagement that makes you an evaluator of the output rather than a passenger of it. This is the step the MIT essay-writers skipped, and the step whose absence showed up on their EEGs.

2 — Prompt with your perspective

Bring your thinking into the prompt: your hypothesis, your uncertainty, your specific question. A prompt carrying your perspective returns output that engages your thinking; a bare task description returns output that replaces it. What you get back is largely determined by how much of yourself you put into the asking.

3 — Evaluate against your engagement

Compare the output to what your own engagement produced. Where it agrees, where it diverges, where it is confident in ways your experience distrusts — the divergences are the actual work, and working them is what builds judgment instead of spending it.

4 — Own the conclusion

Whatever ships under your name, you can defend without reference to the tool — Judge Castel’s standard, applied before the sanction instead of after. If you cannot, you have not finished; you have only accepted.

Category by Category

Writing and communication. Draft yourself first, however rough; then let AI sharpen structure and stress-test the argument. The two-draft discipline is slower by minutes and compounds for decades: over a thousand communications, the professional who keeps the first draft develops as a thinker; the one who skips it produces competent output and quietly stops developing. Same tool, opposite trajectories.

Research and synthesis. Let AI map the terrain; walk the load-bearing ground yourself. The fluency illusion — sounding informed without understanding — is self-concealing, because synthesis papers over gaps so smoothly you cannot feel where they are. The rule from Chapter 10 is absolute: any fact your conclusion stands on, you verify at the source.

Analysis and decisions. Form the hypothesis before running the analysis; interrogate the assumptions after. Use AI to compute, never to decide — the jagged-frontier consultants who fell asleep at the wheel were not lazy, they were confident, and confidence without independent judgment is exactly what nineteen extra percentage points of error looks like.

Creative and ideation. Generate many, select against your taste, transform until recognizable — the full Chapter 9 discipline. The Doshi–Hauser experiment is the standing warning: unfiltered adoption individually helps and collectively homogenizes, and the homogenized middle is now free.

Three Deliverables, Three Sequences

The framework’s honesty is that it scales with stakes, so here it is running at three settings in one working day.

The scheduling email — low stakes, zero judgment content: full delegation, ten seconds, feel nothing. This is the setting most of a workday lives at, and treating it with ceremony would discredit the whole system. The discipline is not applying depth everywhere; it is knowing where not to.

The client analysis — judgment is the product, your name carries it: the full framework, every step. Twenty minutes of position before the prompt. The AI briefed with your perspective and instructed to attack it. The divergences investigated, the load-bearing facts touched at their sources, the conclusion owned to the standard of the courtroom. This is the setting the market-entry memo in Chapter 10 ran at, and it is worth roughly two extra hours against a week of compounding.

The board narrative — highest stakes, most personal: the sequence inverts toward the human. The narrative is drafted entirely by hand first, because a board reads the thinking through the prose and borrowed prose reads borrowed. The AI’s role arrives late and narrow: adversary and editor — find the holes, tighten the argument, say what a hostile director would say. Nothing it produces enters the document unrewritten.

Same person, same tools, three different sequences in one day — and choosing the setting took five seconds each time. That choosing reflex, more than any single technique, is what this chapter installs. It is judgment about where judgment is needed, and like everything in this book, it sharpens with deliberate use.

Scaling It Honestly

The framework is sustainable because it scales with stakes. The routine scheduling email does not need a position brief; hand it over and feel nothing. Reserve the full discipline for work where your judgment is the product, your name carries the consequence, or you will someday defend the reasoning. Knowing quickly which work is which is itself a judgment skill — and like all of them, it sharpens with practice and dulls with delegation.

The Practice

1. For one week, name the step you are on — out loud or in writing — each time you open an AI tool for significant work. The naming is what installs the framework.

2. Identify your weakest category — the one where you most often augment first and accept. Design one specific correction and run it for a month.

3. Once a month, complete one significant task with no AI at all. Not as ideology — as the capability check the endoscopists never scheduled.